Use the canonical Llama model ID

[AnthonyRonning]

Summary

• use llama3-3-70b as the canonical Llama model identifier
• normalize fetched and persisted model IDs so legacy saved chats map to the supported ID
• remove stale legacy aliases and update the remaining examples and reference text

Testing

• just format
• just lint
• just build
• cd frontend && bun test

[coderabbitai]

📝 Walkthrough

Walkthrough

A centralized model ID aliasing system is introduced via a new LLAMA_MODEL_ID constant and MODEL_NAME_ALIASES lookup table. The hardcoded conditional aliasing logic is replaced with this table-driven approach, and duplicate Llama model configurations are consolidated across the codebase.

Changes

Cohort / File(s)	Summary
Aliasing Infrastructure frontend/src/utils/utils.ts	Replaced hardcoded conditional aliasing with centralized MODEL_NAME_ALIASES lookup table; added LLAMA_MODEL_ID constant; updated aliasModelName to use table-based lookup with fallback.
Model Configuration & Deduplication frontend/src/components/ModelSelector.tsx	Updated MODEL_CONFIG to use computed LLAMA_MODEL_ID key instead of hardcoded AWQ identifier; refactored getModelTokenLimit to resolve via aliasModelName; changed deduplication logic to normalize model IDs; removed special-case name-preference filter.
Model State Normalization frontend/src/state/LocalStateContext.tsx	Introduced normalizeAvailableModels function to deduplicate and canonicalize model entries; applied normalization at state initialization and on model list updates.
Configuration & Documentation docs/conversations-api-implementation.md, frontend/public/llms-full.txt, frontend/src/components/apikeys/ProxyConfigSection.tsx	Updated model identifier references from llama-3.3-70b / long AWQ path to standardized llama3-3-70b format; replaced hardcoded strings with LLAMA_MODEL_ID interpolation in example configurations.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• Update Gemma model ID to gemma-3-27b #291 — Modifies MODEL_CONFIG keys and model deduplication logic in ModelSelector.tsx alongside aliasing updates.
• Simplify model selector to Quick/Powerful #432 — Changes model ID aliasing and normalization in both ModelSelector.tsx and LocalStateContext.tsx with overlapping model standardization.
• Remove kimi-k2 model and brain toggle, keep kimi-k2.5 #414 — Updates model aliasing and MODEL_CONFIG legacy model ID mappings.

Poem

🐰 The models are many, the IDs grow long,
So we built a lookup table to set things right and strong!
With LLAMA_MODEL_ID and aliases neat,
No more hardcoded chains—one source of truth, complete! ✨

🚥 Pre-merge checks | ✅ 2

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'Use the canonical Llama model ID' directly and clearly summarizes the main objective of the pull request, which is to standardize on a single canonical Llama model identifier across the codebase.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch canonical-llama3-3-70b

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[devin-ai-integration]

Devin Review found 1 potential issue.

View 4 additional findings in Devin Review.

[devin-ai-integration]

🔴 Dropped backward-compatibility aliases for three old model names

The refactoring from if-else to a lookup table removed three previously-existing aliases:

• "ibnzterrell/Meta-Llama-3.3-70B-Instruct-AWQ-INT4" (was → "llama-3.3-70b", which chains to "llama3-3-70b")
• "qwen3-coder-480b" (was → "kimi-k2-5")
• "leon-se/gemma-3-27b-it-fp8-dynamic" (was → "gemma4-31b")

These aliases exist for backward compatibility with persisted user data. When getChatById loads a chat at LocalStateContext.tsx:241, it calls aliasModelName(parsedChat.model). If a user has an old chat stored with one of these model names, it will now pass through un-aliased. The unrecognized model name then won't match any MODEL_CONFIG entry in ModelSelector.tsx, causing incorrect token limit fallbacks (via getModelTokenLimit at frontend/src/components/ModelSelector.tsx:78) and broken display names. Similarly, persistChat at frontend/src/state/LocalStateContext.tsx:96 will re-persist the unrecognized name rather than migrating it.

Suggested change

	const MODEL_NAME_ALIASES: Record<string, string> = {
	"llama-3.3-70b": LLAMA_MODEL_ID,
	"gemma-3-27b": "gemma4-31b",
	"deepseek-r1-0528": "kimi-k2-5",
	"kimi-k2": "kimi-k2-5",
	"kimi-k2-thinking": "kimi-k2-5"
	};
	const MODEL_NAME_ALIASES: Record<string, string> = {
	"ibnzterrell/Meta-Llama-3.3-70B-Instruct-AWQ-INT4": LLAMA_MODEL_ID,
	"llama-3.3-70b": LLAMA_MODEL_ID,
	"leon-se/gemma-3-27b-it-fp8-dynamic": "gemma4-31b",
	"gemma-3-27b": "gemma4-31b",
	"qwen3-coder-480b": "kimi-k2-5",
	"deepseek-r1-0528": "kimi-k2-5",
	"kimi-k2": "kimi-k2-5",
	"kimi-k2-thinking": "kimi-k2-5"
	};

---

Was this helpful? React with 👍 or 👎 to provide feedback.